Significant slowdown in AI payoff could tip economy into a recession, says Apollo Global's Slok

[00:00:00] Speaker 1: Our next guest warns that a significant slowdown in the payoff of AI could tip the economy into recession. Apollo chief economist and partner, Torsten Schlock, joins us with more. Torsten, great to have you with us. So you're basically asking the question, what if it takes longer for this payoff to actually materialize? Or what if it doesn't happen? What are the odds? How do you assess the risk and the reward here? [00:00:22] Torsten Schlock: Well, let's first agree that AI is making a huge difference in all our lives. So it will be a revolutionary technology. It continues to have a dramatic impact. But the key issue from a stock market perspective, any stock price is essentially the net present value of the future cash flows. So the discounted net present value of future cash flows then suddenly becomes important. What is the slope of those cash flows that are coming in the future? And the question now, of course, is what are the assumptions in market pricing today about how quickly revenues will come to the hyperscalers on the back of the investments that they have made? So it's really this simple observation that a lot of investments are being made and our markets now pricing in that revenues are coming too slowly, too quickly at the right pace. This is essentially the discussion that we're having in markets and the debate, namely, will the revenues come quick enough? [00:01:04] Speaker 3: Yeah. So another debate and Michael Burry of big short fame. It was a little Jenga thing in there. You remember that movie there? It was a great scene. And everyone's like this. I saw the movie. I remember Jenga. Yeah, I remember. This whole notion and he was writing about this. This whole notion that you and we were talking about last night, this depreciation of these high end GPUs. Right. And there's certain marks. And, you know, over the last year, we've seen, you know, some companies like Amazon stretch out that depreciation. And we've seen Meta, for instance, you know, kind of narrow that sort of thing. So we're going to get to a point where the marks are kind of funky. But they are masking some of this performance that we're seeing or they're augmenting, you know, I mean, some of the performance at some point. Don't you think the rubber has to hit the road on this sort of thing? And it's going to be by the hyperscalers. And maybe that's why they trade so poorly, because maybe 2026, or the back half, is going to be the period in which we get more clarity on the marks of these GPUs. [00:01:53] Torsten Schlock: Yeah. The counter argument, of course, is that the demand for compute is basically unlimited. Let's agree that there will be so many different ways that compute will be needed for consumers, for households, and broadly speaking, of course, for companies also. So if that's the case, the question becomes, what is the price that the hyperscalers can get and what is the revenue they can get from that compute? And that's where the conversation becomes so important, in particular with the Chinese models coming in, also the token maxing that you just talked about. All this becomes incredibly important because it, again, brings back this discussion around, well, what if the revenues arrive much faster? Well, then the hyperscalers are actually cheap. Well, if the revenues arrive much slower and the implementation of AI is going to be slower, well, then the revenues, of course, are going to be a lot weaker. So that's why the implicit discussion we should be having is, well, what is the pricing that is in the market today? And what does that assume in terms of consensus expectations? And consensus expectations at the moment are assuming that we basically will have a doubling of the revenue and ultimately of the free cash flow for the hyperscalers in a matter of three, four years. So that's, of course, a very, very optimistic assumption around how much demand there will be and ultimately what price the hyperscalers will be able to charge for that demand. [00:03:01] Speaker 4: So thanks for being here in person. Nice to have you here on the desk. So the other side of this demand, which I believe insatiable is the right quantity of demand, but the supply side response. How do you think about that and what that does to that demand curve and price, to that price curve? [00:03:17] Torsten Schlock: This is very important because the risk, of course, is to this, the demand is essentially unlimited, is that namely that the supply may be a lot slower in rolling out. So that, of course, could imply that the price would ultimately not go down to zero. Instead, there would be a lot of competition because there's so many people who need that compute. So you're right, maybe we have very unlimited demand and we have much more limited supply because there's going to be some challenges and it's going to take some time to roll out all the capacity. Then that would ultimately indeed be a situation where the price of compute would not necessarily be going down that much. And it might even begin in episodes to go higher. [00:03:50] Speaker 5: I dig your energy. I dig your work. And this is sort of going to dovetail into this conversation. A few days ago, the energy shock is over. The rate shock is not over. Exactly. What happens if the rate shock isn't over in this conversation we're having? [00:04:05] Torsten Schlock: Well, what's really, really important is that for a long time, rates have basically been moving up and down with oil prices. So whenever oil prices went up because the straight-up was closed, well, then, of course, rates also went up. But a few weeks ago, something very, very important happened, namely that rates kept on hanging out at higher levels and oil prices really came down a lot. So exactly to your point, the risk now is that interest rates will be higher for longer. And this shift that we've seen in rates markets away from focusing on headline inflation to instead now focusing on core inflation. That's why on the day today when Kevin Walsh announced his task force heads, well, this is, of course, very important because the Fed now needs to deal with that there is less focus on headline inflation. And now there's much more focus on core inflation, which raises this risk, exactly to your point. That rates will be higher for longer. And therefore, the cost of financing will also be higher for longer for anyone who is investing both short-term and long-term. [00:04:52] Speaker 1: So how do you think about what the Fed would do most likely next? I mean, do you think that a cut is more likely than a hike because of economic weakness? [00:05:01] Torsten Schlock: So as we speak, markets are pricing that the Fed will hike twice, once in September and once in March next year. I mean, that's a very, very strong statement from the market saying that there is a shift in attention away from headline inflation maybe coming down. But now we're beginning to worry about core inflation hanging out at higher levels. And for markets, that indeed means that all assets that depend a lot on interest rates, and that is, of course, in particular tech, software, anything that has long-duration cash flows, are going to be more sensitive now that rates are going to stay higher for longer. [00:05:47] Speaker ?: Thank you.